app.py

import os
import torch
import gradio as gr
import numpy as np
from transformers import WhisperProcessor, WhisperForConditionalGeneration
from googletrans import Translator
from gtts import gTTS
import librosa
import tempfile
import soundfile as sf

class RealTimeTranslator:
    def __init__(self):
        # Initialize Whisper model for speech recognition (using tiny model for lower resource usage)
        self.processor = WhisperProcessor.from_pretrained("openai/whisper-tiny")
        self.model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny")

        # Use GPU if available
        self.device = "cuda" if torch.cuda.is_available() else "cpu"
        self.model = self.model.to(self.device)

        # Initialize translator
        self.translator = Translator()

        # Supported languages
        self.languages = {
            'en': 'English',
            'fr': 'French',
            'hi': 'Hindi',
            'es': 'Spanish',
            'de': 'German',
            'ja': 'Japanese'
        }

    def speech_to_text(self, audio_path, source_lang):
        """Convert speech to text using Whisper"""
        try:
            # Load and preprocess audio
            audio, _ = librosa.load(audio_path, sr=16000)
            input_features = self.processor(audio, sampling_rate=16000, return_tensors="pt").input_features
            input_features = input_features.to(self.device)

            # Generate token ids
            predicted_ids = self.model.generate(input_features)

            # Decode token ids to text
            transcription = self.processor.batch_decode(predicted_ids, skip_special_tokens=True)
            return transcription[0]
        except Exception as e:
            return f"Error in speech-to-text: {str(e)}"

    def translate_text(self, text, source_lang, target_lang):
        """Translate text using Google Translate"""
        try:
            translation = self.translator.translate(text, src=source_lang, dest=target_lang)
            return translation.text
        except Exception as e:
            return f"Error in translation: {str(e)}"

    def text_to_speech(self, text, target_lang):
        """Convert text to speech using gTTS"""
        try:
            with tempfile.NamedTemporaryFile(suffix='.wav', delete=False) as fp:
                tts = gTTS(text=text, lang=target_lang)
                tts.save(fp.name)
                return fp.name
        except Exception as e:
            return f"Error in text-to-speech: {str(e)}"

    def process_audio(self, audio, source_lang, target_lang):
        """Complete pipeline: Speech → Text → Translation → Speech"""
        try:
            if audio is None:
                return None, "No audio input received", "Please provide audio input"

            # Save input audio temporarily
            with tempfile.NamedTemporaryFile(suffix='.wav', delete=False) as fp:
                sf.write(fp.name, audio[1], audio[0])
                audio_path = fp.name

            # Speech to text
            text = self.speech_to_text(audio_path, source_lang)
            if "Error" in text:
                return None, text, ""

            # Translate text
            translated_text = self.translate_text(text, source_lang, target_lang)
            if "Error" in translated_text:
                return None, text, translated_text

            # Text to speech
            output_audio_path = self.text_to_speech(translated_text, target_lang)
            if "Error" in output_audio_path:
                return None, text, translated_text

            # Load the generated audio
            output_audio, sr = librosa.load(output_audio_path)

            # Clean up temporary files
            os.unlink(audio_path)
            os.unlink(output_audio_path)

            return (sr, output_audio), text, translated_text

        except Exception as e:
            return None, f"Error: {str(e)}", f"Error: {str(e)}"

def create_gradio_interface():
    translator = RealTimeTranslator()

    # Create the Gradio interface
    demo = gr.Interface(
        fn=translator.process_audio,
        inputs=[
            gr.Audio(sources=["microphone"], type="numpy", label="Input Audio"),
            gr.Dropdown(choices=list(translator.languages.keys()), value="en", label="Source Language"),
            gr.Dropdown(choices=list(translator.languages.keys()), value="fr", label="Target Language")
        ],
        outputs=[
            gr.Audio(label="Translated Audio"),
            gr.Textbox(label="Original Text"),
            gr.Textbox(label="Translated Text")
        ],
        title="Real-time Language Translator",
        description="Speak in your language and get instant translation in the target language. Please ensure your device is set to speakerphone mode for best results.",
        examples=[
            [None, "en", "fr"],
            [None, "hi", "en"],
            [None, "es", "ja"]
        ]
    )
    return demo

if __name__ == "__main__":
    demo = create_gradio_interface()
    demo.launch(share=True, debug=True)